---
title: "AFL KPI Analysis Dashboard"
output: 
  flexdashboard::flex_dashboard:
    storyboard: true
    theme: united  
    source_code: embed
    social: ["menu"]
---

<style>
  .custom-header {
    position: relative;
    padding-top: 10px;
    padding-bottom: 10px;
    margin-bottom: 20px;
    border-bottom: 2px solid #ddd;
  }

  .custom-header img {
    position: absolute;
    top: 0;
    right: 0;
    height: 60px;
  }
</style>

<div class="custom-header">
  <h2>AFL Regression Models: 2012 KPI Analysis</h2>
</div>

```{r setup, include=FALSE}
# Load Packages
library(tidyverse)
library(plotly)
library(flexdashboard)
library(dplyr)
library(readxl)
library(DT)
library(reshape2)
library(RColorBrewer)
library(glmnet)
library(knitr)

# Load dataset 
afl <- read_excel("/home/gthornton1999/VU/2. Introduction/Assignment 3/Data/Football dataset.xlsx")
```

### Data Table


```{r Interactive Table}

# Create a value for KPI column names
kpi_cols <- c("Metres_Gained", "Uncontested_Possessions", "Contested_Possessions", "Contested_Marks", "Uncontested_Marks", "HB_Receives", "Clearances", "TimeInPossession_Differential%", "Clanger_Kicks", "HB", "HB_Efficiency%", "Kick", "Kick_HB_Ratio%", "Kick_Efficiency%", "Tackle", "Turnover", "Behind", "Goal", "Score", "Score_Accuracy%", "Mark", "Mark_PlayOn", "Inside_50", "Rebound_50")

# Create a data frame summing the quarterly differentials into match form
match_df <- afl %>%
  group_by(Season, Round, Team, Final_Ladder) %>%
  summarise(across(where(is.numeric), sum, na.rm = TRUE),
            Final_Ladder = first(Final_Ladder)) %>%
  ungroup()
# Aggregate the round KPIs to season long KPIs
season_df <- match_df %>%
  group_by(Season, Team, Final_Ladder) %>%
  summarise(across(all_of(kpi_cols), sum), .groups = "drop")

# rounded data set to 1 dp
season_df_rounded <- season_df %>% 
  mutate(across(where(is.numeric), ~ round(.x, 1)))



# create 
datatable(season_df_rounded, 
          extensions = 'Buttons',
          options = list(
            dom = 'Bfrtip',
            buttons = c('copy', 'csv', 'excel', 'pdf', 'print'),
            paging = FALSE,
            scrolly = "500px",
            scrollCollapse = TRUE
          ),
          rownames = FALSE,
          class = 'stripe hover'
          )

```





### Regression Techniques

```{r Lasso Regression}

# Correlation matrix of KPIs
cor_matrix <- cor(season_df %>% select(-Season, -Team), use = "complete.obs")

# Ranked list of KPIs that directly effect a low ladder position (the most influential KPIs)
cor_with_ladder <- cor_matrix["Final_Ladder", ]
cor_with_ladder <- sort(cor_with_ladder[-1])  # remove self-correlation and sort

# Table of the ranked KPIs
knitr::kable(cor_with_ladder)

# Heatmap of correlations 
library(reshape2)
library(RColorBrewer) # maybe virirdis??

# correlation matrix for KPIs
cor_matrix <- cor(season_df %>% select(-Season, -Team), use = "complete.obs")
# melt correlation matrix for gg plot 
# melt = reshapes the data into long format for plotting
cor_melt <- melt(cor_matrix)
# Add correlation with Final_Ladder to the melted data
cor_ladder_df <- data.frame(KPI = names(cor_with_ladder), Correlation = cor_with_ladder)
cor_melt$Final_Ladder <- cor_ladder_df$Correlation[match(cor_melt$Var1, cor_ladder_df$KPI)]
#plot heatmap
plot_ly(
  data = cor_melt, 
  x = ~Var1, 
  y = ~Var2, 
  z = ~value, 
  type = "heatmap", 
  colors = c("orange", "white", "purple"), 
  text = ~paste("KPI1: ", Var1, "<br>KPI2: ", Var2, "<br>Correlation:", round(value, 2)),  # Hover text of KPI v KPI and correlation
  hoverinfo = "text"  # Show text on hover
) %>%
  layout(
    title = "Interactive Heatmap of KPIs and Final Ladder Position",
    xaxis = list(title = "KPI", tickangle = 45),
    yaxis = list(title = "KPI")
  )


# a matrix of predictors
X <- model.matrix(Final_Ladder ~ . -Season -Team, data = season_df)[, -1]
# a numeric outcome value
y <- season_df$Final_Ladder

lasso_model <- cv.glmnet(X, y, alpha = 1) 
#plot(lasso_model)

coef(lasso_model, s = "lambda.min") 


# Create neater table
coef_df <- as.matrix(coef(lasso_model, s = "lambda.min"))
coef_df <- data.frame(KPI = rownames(coef_df), Coefficient = coef_df[,1])
coef_df <- coef_df[coef_df$Coefficient != 0, ]  # remove zeroed-out KPIs

knitr::kable(coef_df)

# Predicted Ladder position
# assess accuracy
intercept <- 9.5
goal_coef <- -0.04
# calculate predicted ladder position using lasso equation
season_df$Predicted_Ladder_Position <- intercept + goal_coef * season_df$Goal
# Create a comparison table
ladder_comparison <- season_df[, c("Team", "Goal", "Predicted_Ladder_Position", "Final_Ladder")]
# calculate r2 value for accuracy.
r_squared <- cor(season_df$Final_Ladder, season_df$Predicted_Ladder_Position)^2
# kable table with accuracy
kable(ladder_comparison, caption = paste("Predicted vs Actual Ladder Positions Using the Lasso Regression Equation (R² =", round(r_squared, 3), ")"))
```




### Key Points for Staff

```{r Interactive plot of data}

# plot of goal vs ladder position. 
GoalvLadder <- ggplot(season_df, aes(x = Goal, y = Final_Ladder)) +
  geom_point(aes(colour = Team)) +
  geom_smooth(method = "lm") +
  labs(title = "Goal Differential vs Final Ladder Position",
       x = "Goal Differential",
       y = "Final Ladder Position") +
  theme_classic()
# interactive
ggplotly(GoalvLadder)

```